Getting this error when picking a image

[Camekan]

🔎 Search before asking

• I have searched the PaddleOCR Docs and found no similar bug report.
• I have searched the PaddleOCR Issues and found no similar bug report.
• I have searched the PaddleOCR Discussions and found no similar bug report.

🐛 Bug (问题描述)

I`m using it for OCR on korean manhwas. OCR gets done by PaddleOCR and gets sent to the AI model as text to get translated. Getting a error like this: I0123 10:21:47.322901 16792 onednn_context.cc:81] oneDNN v3.6.2
🔥 Error: (Unimplemented) ConvertPirAttribute2RuntimeAttribute not support [pir::ArrayAttributepir::DoubleAttribute] (at ..\paddle\fluid\framework\new_executor\instruction\onednn\onednn_instruction.cc:118)

🏃‍♂️ Environment (运行环境)

Windows 10
Python 3.10.11
RX 590 8 GB
16 GB RAM
Ryzen 5 5600
Using Vulkan
Using manga_server.py provided below
Using Manga Translator extension for Firefox below

manga_server.py

Manga translator.zip

🌰 Minimal Reproducible Example (最小可复现问题的Demo)

Getting the error using it with my Firefox add-on. When I pick an area of a korean manhwa text like image it gives the error: I0123 10:21:47.322901 16792 onednn_context.cc:81] oneDNN v3.6.2
🔥 Error: (Unimplemented) ConvertPirAttribute2RuntimeAttribute not support [pir::ArrayAttributepir::DoubleAttribute] (at ..\paddle\fluid\framework\new_executor\instruction\onednn\onednn_instruction.cc:118)

[arthurbrenno]

Same.

[Camekan]

Forgot to mention I solved it. Kept asking Gemini, Claude. You can find out how I solved from these chats: https://gemini.google.com/share/b312d77d53c2 and https://gemini.google.com/share/796fef101cdf

Btw, had to drop my paddleocr version to 2.7.0.3 and paddlepaddle to 2.6.2 for my specific script. Dont know if it would help your case but try dropping versions like first your paddleocr and then other things if you know your things or if you don`t, again, ask AIs.

Or, If you don`t wanna scroll through them here is an explanation by Gemini:

---

We encountered this issue while running PaddleOCR on Windows/CPU. The error ConvertPirAttribute2RuntimeAttribute is caused by a conflict between the new PIR (Program Intermediate Representation) executor introduced in PaddlePaddle v2.6+ and the oneDNN (MKLDNN) accelerator.

Even if you set os.environ['FLAGS_use_mkldnn'] = '0', the PaddleOCR class may still default to enabling MKLDNN internally during initialization, triggering the crash.

The Solution: You must explicitly disable MKLDNN when initializing the PaddleOCR class in your Python script.

Change this:
ocr = PaddleOCR(use_angle_cls=True, lang='korean', use_gpu=False)

To this:

Explicitly pass enable_mkldnn=False to bypass the buggy PIR conversion

ocr = PaddleOCR(use_angle_cls=True, lang='korean', use_gpu=False, enable_mkldnn=False)

This forces Paddle to use the standard CPU kernels instead of the optimized oneDNN kernels, which bypasses the missing attribute conversion that causes the crash.

Basically, if you have the same error just use these or if you dont know how to add these, ask an AI to add it to your script or whatever. I added the PaddleOCR to my manga/manhwa translation script, which btw built by the Gemini and refined by Claude and other AIs. Thank god for the people who creates those AIs especially Claude it is really smart. I wouldn`t know how to write a line of code if I had a gun to my head really.

[Camekan]

Just ask AIs or Claude especially since it is better from my observations at coding. Claude gives you a very good explanation if you send your code, error, how you encountered it. Ask it to find out the solution from "websites like stackoverflow, reddit, github, huggingface". It gives you a better solution compared to randomly trying to fix your code.

[344089386]

paddle:3.3.0
paddleocr:3.3.0
cpu: AMD

I have also encountered this problem.
Let's try it at a lower level first.

[lxf9501]

Same

[Camekan]

Looks like it`s a problem on AMD CPUs or CPUs in general. Either drop your versions: paddleocr version to 2.7.0.3 and paddlepaddle to 2.6.2. Or, use paddleocr 3.4.0 and paddlepaddle 3.2.1. Right now, I can use both PP-OCRv5 and PaddleOCR-VL-1.5 with those versions. This is my script if anyone wants to take a look at it for solutions. I also uploaded a txt file for an explanation by Gemini who made it work for me. Explanation includes what made my script work if you wanna take a look at it.

https://gemini.google.com/share/a58b3a5d1a3c

manga.py

b.txt

[TMS-Namespace]

Downgrading paddlepaddle to just 3.2.2 fixed the issue for me, while using paddleocr>=3.3.2.

[NandaYanxr]

Downgrading paddlepaddle to just 3.2.2 fixed the issue for me, while using paddleocr>=3.3.2.

TMS's solution works for me as well

[344089386]

Downgrading to just fixed the issue for me, while using .paddlepaddle``3.2.2``paddleocr>=3.3.2

TMS's solution works for me as well，good

[makdinoven]

Forgot to mention I solved it. Kept asking Gemini, Claude. You can find out how I solved from these chats: https://gemini.google.com/share/b312d77d53c2 and https://gemini.google.com/share/796fef101cdf

Btw, had to drop my paddleocr version to 2.7.0.3 and paddlepaddle to 2.6.2 for my specific script. Dont know if it would help your case but try dropping versions like first your paddleocr and then other things if you know your things or if you don`t, again, ask AIs.

Or, If you don`t wanna scroll through them here is an explanation by Gemini:

We encountered this issue while running PaddleOCR on Windows/CPU. The error ConvertPirAttribute2RuntimeAttribute is caused by a conflict between the new PIR (Program Intermediate Representation) executor introduced in PaddlePaddle v2.6+ and the oneDNN (MKLDNN) accelerator.

Even if you set os.environ['FLAGS_use_mkldnn'] = '0', the PaddleOCR class may still default to enabling MKLDNN internally during initialization, triggering the crash.

The Solution: You must explicitly disable MKLDNN when initializing the PaddleOCR class in your Python script.

Change this: ocr = PaddleOCR(use_angle_cls=True, lang='korean', use_gpu=False)

To this:

Explicitly pass enable_mkldnn=False to bypass the buggy PIR conversion

ocr = PaddleOCR(use_angle_cls=True, lang='korean', use_gpu=False, enable_mkldnn=False)

This forces Paddle to use the standard CPU kernels instead of the optimized oneDNN kernels, which bypasses the missing attribute conversion that causes the crash.

Basically, if you have the same error just use these or if you dont know how to add these, ask an AI to add it to your script or whatever. I added the PaddleOCR to my manga/manhwa translation script, which btw built by the Gemini and refined by Claude and other AIs. Thank god for the people who creates those AIs especially Claude it is really smart. I wouldn`t know how to write a line of code if I had a gun to my head really.

PaddleOCR 3.4.0 + PaddlePaddle 3.3.0: MKLDNN Backend Crash

Symptoms

When running PaddleOCR inference on CPU, the process crashes with:

NotImplementedError: (Unimplemented) ConvertPirAttribute2RuntimeAttribute not support
[pir::ArrayAttribute<pir::DoubleAttribute>]
(at /paddle/paddle/fluid/framework/new_executor/instruction/onednn/onednn_instruction.cc:116)

Other variations of the same root cause:

• ResourceExhaustedError: Fail to alloc memory of 24546115584 size
• PaddleOCR.predict() got an unexpected keyword argument 'cls'

Root Cause

PaddleOCR 3.4.0 uses PaddleX as its inference backend. For CPU inference, PaddleX defaults to run_mode="mkldnn" (Intel oneDNN). You can verify this yourself:

from paddlex.inference.models.common.static_infer import get_default_run_mode

print(get_default_run_mode("PP-OCRv4_mobile_det", "cpu"))  # "mkldnn"
print(get_default_run_mode("en_PP-OCRv4_mobile_rec", "cpu"))  # "mkldnn"

PaddlePaddle 3.3.0 has a bug where its new PIR (Paddle Intermediate Representation) is incompatible with the MKLDNN instruction converter. The crash happens deep inside the static inference engine at predictor.run().

Important: Setting FLAGS_use_mkldnn=0 as an environment variable does not help. PaddleX controls the inference backend through its own PaddlePredictorOption.run_mode and ignores the Paddle-level flag.

Fix

Pass enable_mkldnn=False to the PaddleOCR() constructor:

from paddleocr import PaddleOCR

ocr = PaddleOCR(
    lang="en",
    ocr_version="PP-OCRv4",
    enable_mkldnn=False,  # prevents MKLDNN/PIR crash
)

results = list(ocr.predict(image_array))

How it works

The enable_mkldnn kwarg flows through the following chain:

1. PaddleOCR.__init__(**kwargs) passes unknown kwargs to super().__init__()
2. PaddleXPipelineWrapper.__init__() calls parse_common_args(kwargs)
3. parse_common_args() recognizes enable_mkldnn and stores it in common config
4. PaddleX configures the inference predictor with run_mode="paddle" instead of run_mode="mkldnn"

This bypasses the broken MKLDNN/PIR code path entirely.

Docker Compose

As additional insurance, set the environment variable in your docker-compose.yml:

services:
  worker:
    environment:
      - FLAGS_use_mkldnn=0
      - PADDLE_PDX_DISABLE_MODEL_SOURCE_CHECK=True

The second variable skips a slow connectivity check to HuggingFace on startup.

Affected Versions

• PaddleOCR 3.4.0
• PaddlePaddle 3.3.0
• CPU-only inference on x86_64 Linux

GPU inference is not affected since it uses CUDA, not MKLDNN.

[Camekan]

Btw, it was unbelieveably slow waiting for PaddleOCR to do its thing with VL 1.5 version. Just switched to Qwen3-VL-8B-Instruct at full quantization with kobold.cpp. It picks up korean manhwas better than VL 1.5 and translates them at the same time not making me wait 2 minutes.